Visual Search is making images Shoppable

Pricesearcher presented at the Retail Week Tech event in September and during the event there were many interesting panels and presentations and so on this blog we share information regarding visual search and its growth:

  • Visual search gives consumers the power to make images shoppable, to search for similar items and to search for things in ways they simply can’t describe in words
  • It also has wider business uses in terms of analysis of product sales and subsequent forecasting of future orders
  • The technology is not yet plug and play due to its complexity and the time required for an algorithm to learn
  • In order to get meaningful, relevant results for the consumer it requires a very deep pool of images: 3 – 4 million images is the minimum
 Girl Taking Photo With Smartphone On Sunny Day

The Future of Visual Search



The traditional text keyword search typed into a search engine is being augmented by voice search (Digital Assistants like Alexa, Cortana, Google Assistant, Bixby or Siri) and image search. The thinking is that text search is not going to be replaced but rather where it’s more convenient to use voice search (when driving for example) or more convenient to use image search (when you’ve seen an item of clothing you like and managed to take a quick photo) then people will use the method that is easiest and most reliable. Incremental change is already underway……..

What are the uses of Visual Search for Consumers?

The technology has already been developed and it gives consumers the ability to do several things:

  1. Image search – Take a picture on an item on your smartphone and then search for that product using the photo 
  2. ‘Shop the look” – within sites with the functionality exists you can hover over an image  of a product and then image search to see similar listings within the site
  3. Chatbot on FB messenger – sending images via messenger to then get shoppable results.

Consumers want to take an image and then shop based on that image search so its a customer acquisition tool with additional benefits in terms of increased conversion and basket size.

Other non-consumer facing applications include:

  1. Analysis – What products sold well by image? Traditional analysis on the season’s best lines have traditionally been done at the SKU level but what if there was a general trend for a certain material e.g. leather or a particular style e.g. skinny jeans that cut across SKU’s in a way that was impossible to pull out from traditional analysis. Visual search analysis may be able to offer much more powerful insights to then infer future buying
  2. Forecasting – understanding trends in the wider market based on images that are prevalent across social and online media. On an aggregated basis this level of information could also guide future buying decisions and ultimately make fashion trends self-fulfilling if buying departments follow what leading fashion icons are already wearing

Image search is a very complex form of text search

Accuracy is key to visual search and it’s important to think about what images are to a computer. Whether a search is made by text, image or voice it is still just data and its still broken down into its component parts of 0’s and 1’s. Images are turned into very long and complicated text. There are many particular aspects which make image search so technically difficult:

  1. Which is the relevant part of the image? How does a computer zone in on the t-shirt that you’re interested in if there is also a pair of trousers, belt, sunglasses and a car in the background? eBay has tackled this by building a cropping functionality into their visual search as a standard part of the process to aid accuracy. 
  2. Algorithm challenge with different materials. Particularly with clothing and apparel the material used is of the utmost importance and yet its very difficult for a computer to ascertain whether a material is wool or cashmere for example.
  3. You need 3-4 million images in a data set to enable meaningful search. In order to generate relevant results a very large data set is required – into the millions certainly not the tens or even hundreds of thousands.
  4. You need to train an algorithm so it can learn. This technology is not yet at the plug and play stage; its very complex with even the leading fashion e-commerce sites using specialist outsource tech providers and working with them on an on-going basis. Computer vision is still very much an evolving field of science. 
Everyone is familiar with the music app Shazam which recognises every song but this is actually much easier for songs because each one is unique and so a computer can then do a True:False test . Its actually much harder to find ”similar” songs and the same is also true when it comes to finding similar images because its not a Yes:No, 1:0, binary option which a computer can easily perform.


In the year that the word ”Emoji” has entered the English dictionary, it’s clear that the text search is being augmented by images as a form of communication and its expected that visual search will slowly become part of our lives. The technology is now delivering and visual search cuts across language barriers making it a very natural process. An image can describe things in ways consumers can’t describe or even conceive in their imagination. Looking for similar items to very expensive fashion/home products is becoming easier and the proliferation of ideas and fashion into e-commerce will only become faster.

Credit to panellist Dr Eduard Vazquez, Cortexica

Click here to find out more about Pricesearcher and our mission to index all the world’s prices

About Ben Morgan

Head of Commercial for Pricesearcher

1 Response

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s